Re: skbuff panic

[Oliver Hartkopp <socketcan@hartkopp.net>]

On 06.07.2014 07:07, Austin Schuh wrote:

> What makes you think that someone opened a PF_PACKET socket?  I'm
> curious, since that observation may help me produce a more reliable
> test case and debug it some more myself.

If you look in your call trace is says:

skb_push+0x38/0x39          (-->> which panics)
packet_rcv_spkt+0x98/0xdf
__netif_receive_skb_core+0x459/0x4dc
get_parent_ip+0xe/0x3e
__netif_receive_skb+0x53/0x65

As packet_rcv_spkt() is located in net/packet/af_packet.c there must be some
user at this early stage of system boot to make PF_PACKET process the CAN frame.

In https://gitorious.org/linux-can/can-tests there's a tst-packet.c program
which uses the PF_PACKET socket to send/receive CAN frames.

I was using tst-packet.c about four years ago for a test - without any
problems. But maybe something in the network layer changed, so that CAN frame
skbs need to be created with a different setup now.

I'll try ASAP if tst-packet.c still works as expected on my machine.

> 
> I'm going to work on reproducing the panic more reliable, and then
> I'll give your patch a whirl.  Currently, only PCs that are on the
> other end of cell modems out in the field seem to be triggering the
> panic.  I'm hesitant to do excessive experimentation on something that
> takes a plane trip to fix.  That just means that this will take longer
> to debug that I'd like...

Indeed this should not be the plan ...
Let's save the world by saving carbon dioxide ;-)

Best regards,
Oliver